[lwip-users] Memory leak due to synthesized DDoS

[Stephen Cowell]

My product is a modbus TCPIP board, using the Atmel SAM4E16E based off 
of the SAM4E-EK.  It is using LWiP 1.4.1 with pbuf.c and pbuf.h from 
version 2.1.3.  I know... cringe... but I did this recently, during 
troubleshooting... helped some, see below.

I'm banging it real hard using DaqFactory to create two simultaneous 
threads sending modbus requests to the same port/address.  As I do this 
I can see the pbuf_pool pbufs being created higher and higher up the 
pool... I can debug on the pbuf deallocation and see that the ->next is 
NULL, so there is no connection to the rest of the chain.  If I let this 
go on for long enough the processor will hard fault.

I had this behavior happen very quickly before I moved from LWiP mem 
management to GNU C library malloc()... now it takes a long time to 
die.  Also, after I spliced in the pbuf code from the latest-greatest it 
did seem to become more robust... but I can still watch the allocated 
pbufs climb up the pool, leaving behind orphan pbufs as a memory leak.

I guess my main question is... should LWiP be able to survive this kind 
of abuse?  I'm able to hit it with requests about every two 
milliseconds... and it takes about an hour before it runs out of 
memory.  Would I notice any improvement by completing the port to 2.1.3 
(a significant effort)?  Is LWiP tested in this way?  It's worth noting 
that DaqFactory is glitching the inputs it receives sometimes... but 
Wireshark shows them to be rock solid (when there's not a timeout or 
port locked complaint).  Am I abusing it too hard? Thanks for your 
time... Steve.